LLM Junior

I've been using iTunes for 25 years. It is a terribly bad piece of software, but one of only a few options to manage a local library of music files with somewhat universal interoperability. In particular the playlists created on an iPod transfer to other Apple (and Windows) computers as well as are recognized by independent local media management and streaming systems like the Sqeezebox Logitech Lyrion.

Over the years iTunes has undergone some minor changes (the latest one is nearly as bad as the original), and some of them have somehow (in my case) resulted in the music database becoming corrupted. Long story short, the only way out was to restore the backed-up mp3 files and let iTunes rescan the entire library. The playlists were unfortunately lost this way, as neither the mp3s nor the file folder structure have any information about the playlist metadata. iTunes itself have since moved the root library folder from iTunes Music to iTunes Media (and also some other changes), so restoring playlists from the backed up library XML file did not work (the XML has path pointers to the actual files and when the file location changes, iTunes itself cannot figure out which files are pointed to).

The task then was to "fix" the library XML by regenerating the path pointers. I tried this on my own but got bogged down searching for the definition of the iTunes XML file format. Most answers were "it is an XML format" which was useless.

Hence came the idea of trying an LLM:

after reorganizing the iTunes library on my windows computer, i lost playlists. can you take a look a the attached files and import the playlists from the "Old Library.xml" to the "New Library.xml" please?

Perplexity gave up citing too many tokens (I have about 25 thousand music files). Chat GPT was nicer (after I upgraded to the Plus version):

Yes — please upload both Old Library.xml and New Library.xml here, and I can extract the playlists from the old file and merge them into the new one so you get them back.

It contemplated a bit and reported:

I merged the playlists from your old library into the new one, skipping system playlists and avoiding duplicates (based on playlist persistent ID or name). I also remapped each track reference to the tracks in your new library using Persistent ID first, then file location, then basic metadata.

It even offered me the Python script to do (or repeat) the merge locally.

Long story short - it worked. The playlists are live again.

The second exercise was writing an emulator of an emergency light fixture that implements a simple query - response protocol over UART. I uploaded the UART protocol definition PDF telling ChatGPT to "generate a sample code for Arduino to emulate an emergency luminaire that implements this protocol". Apart from generating the code it also provided useful comments on the source document, like being more precise on endianness etc.

This is about what can be expected from a junior developer. And people in the industry confirm that - to the point that they (the seniors) often prefer working with LLMs than with human juniors. The grand effect is that it is not that easy now to find a job as a junior developer. Which in most cases is the prerequisite to become a regular and then a senior.

The career chain seems to be broken, or at least starts high enough that many are unable to reach out to it. This sounds like a big problem in the grand scheme of things. 

P.S. Having such tasks automated by a machine causes many to fall into the "AI" trap. I have carefully avoided the term here, as we must remember we are dealing with language models, the statistical text output engines. They seem to be intelligent because they are eloquent, but are not able to design new things. They are just very good at rephrasing what has already been done by humans. Although that alone can be a big productivity boost if humans know how to use the tools.